基于GRA的PSO-BP神经网络斜轧穿孔管形预测  

Prediction of pipe shape in cross-rolled piercing by PSO-BP neural network based on GRA

作　　者：王清华[1] 加世滢 胡建华[2] 双远华[2] 赵铁琳 Wang Qinghua;Jia Shiying;Hu Jianhua;Shuang Yuanhua;Zhao Tielin(School of Electronic and Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China;School of Materials Science and Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China;Technical Center of Taiyuan Heavy Industry Co.,Ltd.,Taiyuan 030024,China)

机构地区：[1]太原科技大学电子信息工程学院,山西太原030024 [2]太原科技大学材料科学与工程学院,山西太原030024 [3]太原重工股份有限公司技术中心,山西太原030024

出　　处：《锻压技术》2022年第8期88-94,共7页Forging & Stamping Technology

基　　金：山西省科技重大专项(20191102009)。

摘　　要：针对斜轧穿孔中无缝钢管管形计算复杂且精度不高的缺陷,提出了基于灰色关联度分析(GRA)的PSO-BP神经网络管形预测模型。由于轧制过程中影响管形的因素较多,通过灰色关联度分析对工艺参数进行了相关性分析,选择相关度较高的影响因素作为输入;并使用粒子群优化算法对BP神经网络进行优化,确定了最佳的神经网络结构,构建了无缝钢管的斜轧穿孔管形预测模型。最后,应用现场数据对该模型进行了训练和测试,并将其与BP神经网络和传统数学模型进行了对比分析。研究结果表明:该预测模型的精度较高、可靠性较好,为提高无缝钢管的生产质量奠定了基础。For the defects of complicated calculation and low precision for pipe shape of seamless steel pipe in cross-rolled piercing,a pipe shape prediction model of PSO-BP neural network based on grey relational analysis was proposed.Since there were many factors affecting the pipe shape during the rolling process,the correlation analysis of the process parameters was carried out by the gray correlation analysis,and the influencing factors with higher correlation were selected as the input.Then,the BP neural network was optimized by particle swarm optimization algorithm,the optimal neural network structure was determined,and the pipe shape prediction model of crossrolled piercing for seamless steel pipe was constructed.Finally,the model was trained and tested by field data and was compared with BP neural network and traditional mathe-matical models.The research results show that the prediction model has higher accuracy and better reliability,which lays a foundation for improving the production quality of seamless steel pipes.

关 键 词：斜轧穿孔 管形预测 灰色关联度分析 粒子群优化算法 BP神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]